CCTV (中央电视台) 3.15曝光“AI投毒”：GEO成新赛道，机遇与风险并存

CCTV exposé and the rise of GEO

CCTV (中央电视台) 3.15晚会曝光了所谓的AI“投毒”现象——一些品牌 reportedly 通过技术和内容布局，让生成式AI在回答时优先提到自己，从而获得“被AI说出来”的曝光。TMTPost reports this practice is rapidly professionalizing as a service industry under the name Generative Engine Optimization (GEO，生成式引擎优化). What was once search-engine optimization for clicks is being repackaged for AI answers. Who benefits? Brands chasing a new information gateway — and the firms that sell them the tactic.

How GEO works, according to practitioners

Industry practitioners interviewed for the report outlined a simple playbook: make content that large language models can easily parse and cite. They summarized it as “DDS”: Semantic Depth, Data Support, Authoritative Source. Tactics combine content work (FAQ-style pages, clear conclusions, citations) with technical plumbing (Schema markup, “machine‑readable” versions of sites). Domestic vendors such as BlueFocus (蓝色光标) and ZheWen Hulian (浙文互联) have launched GEO products; it has been reported that startup Profound has raised multiple rounds and claimed valuation milestones as investors pile in. The goal is not clicks but appearing in the AI-generated answer itself.

Market dynamics and geopolitical context

The commercial logic is straightforward: as users shift from search pages to chat-based AI, attention and conversion funnels change. It has been reported that traditional search traffic may decline while AI assistant usage soars; figures cited in the piece — for example, high user numbers for ChatGPT and Doubao (豆包) — should be read as reported estimates rather than independently verified facts. GEO vendors price services from low-cost keyword packages to high-end technical deployments. At the same time, geopolitics and regulation matter: export controls on advanced chips, growing scrutiny of AI transparency in the U.S. and EU, and platform governance all influence which optimization tactics survive. And when platforms control both the model and the distribution, brands can push — but platforms ultimately decide.

The unresolved risks

The CCTV report and practitioners flag several unresolved problems. Measurement is hard: if users get answers inside the chat and never click, how do you attribute conversion? Model outputs are non‑deterministic; the same prompt can yield different answers at different times. Mass “scale-up” content deployments may amount to what critics call AI polling or “poisoning,” degrading signal quality for everyone. Who enforces quality? Platforms, regulators or the market? It has been reported that some GEO practices already resemble low‑quality, high‑volume content mills — a dynamic likely to draw regulatory attention as governments push for greater AI transparency and accountability. GEO is already a business; whether it becomes a sustainable, ethical industry is still an open question.